Fire Detection Algorithms Using Multimodal ... - Bilkent University

CHAPTER 2. FLAME DETECTION IN VISIBLE RANGE VIDEO 31moving objects causes Method-2.3 to give false alarms in Movies 1, 3, 7 and9. If Method-2.1 is used, moving fire-colored ordinary objects do not cause analarm to be raised. This is because the cyclic movement of flames is taken intoaccount in our method, as well as the spatial variation in the color/brightnessvalues of the moving fire-colored regions. Method-2.1 successfully detects fire invideos covering various scenarios, including partial occlusion of the flame. Sampleimages showing the detected regions are presented in Fig. 2.11.In Movie 11, a man wearing a fire-colored shirt intentionally waves his arms tomimic the quasi-periodic flicker behavior in flames. Although all of the methodsproduce false alarms in this Movie, Method 1 significantly decreases the numberof false positives relative to Methods-2.2 and 2.3.These methods are also compared to each other in terms of computationalcost (as shown in Table 2.3). Movies in Tables 2.2 and 2.3 are all captured at10 fps with a frame size of 320 by 240 pixels. The average processing timesper frame are 17.0 msec, 12.5 msec and 14.5 msec, for our method, Method-2.2,and Method-2.3, respectively. Our method is computationally more demandingdue to additional wavelet analysis based steps. Since only shift and add typeoperations take place when convolving signals with the wavelet filters, additionalcost is not high. Our implementation works in real-time for videos with framesize 320 by 240 pixels, captured at 10 fps or higher in a PC.The video clips that we tested our method contain a total of 83,745 frames in61 sequences. In 19 of the sequences fire takes place. Our method is successful indetecting fire in all of these sequences. This corresponds to a fire detection rateof 1.0. A fire contour recognition rate of 0.999 is reported in [48], which correspondsto a fire detection rate of 0.999. Our overall false alarm (false positive)rate is 0.001. It is reported that non-fire contour recognition rate is 1.0 in [48]which corresponds to a false alarm rate of 0. The video sequences containing firein [48] are not publicly available. Therefore we used our own data set. We alsotest our method with the data set of the EC funded Context Aware Vision usingImage-based Active Recognition (CAVIAR) project [8], publicly available atURL: http://homepages.inf.ed.ac.uk/rbf/CAVIAR/. Although there are a lot of